european university at st. petersburg the method of quantile regression, a new approach to actuarial...

13
European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev Saint Petersburg State University

Post on 21-Dec-2015

224 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

European University at St. Petersburg

THE METHOD OF QUANTILE REGRESSION,

A NEW APPROACH TO ACTUARIAL MATHEMATICS

Authors: Ruslan Abduramanov

Andrey Kudryavtsev

Saint Petersburg State University

Page 2: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

INTRODUCTION

How correctly estimate a net premium?

Classical approaches:• An expected net premium.

• Risk loadings.

• Premium principles.

Disadvantages of classical approaches:• The problem of outliers.

• Fitting the theoretical distribution function.

• Independence of losses.

Page 3: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

Advantages of the method of quantile regression:• In a series of observations, there is a small portion of

“outliers”.

• The distribution of the sample can be accurately fitted.

• Observations from the sample are not independent.

General model of quantile regression:

QUANTILE REGRESSION

'' :

'

:

' 11

miniiii xyi

iixyi

ii xyxyn

Page 4: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

Quantile regression,

as a linear programming problem:

Several important consequences:• It is guaranteed that the computing procedure will be

finished with a finite number of iterations.

• Without changing estimates of s, can be increased/decreased to , robust.

QUANTILE REGRESSION

othercase

xyxyu

othercase

xyxyu

where

u

u

yuuX

uu

iiiii

iiiii

,0

,,0

,

,

0

0

min111

iy

/

Page 5: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

-level of significance, then using well known quantile approach to rating:

where

In this case

So problem is to estimate a conditional distribution function

A NEW APPROACH TO ESTIMATENET PREMIUM RATES

,)()(:inf 1 XX FxxFxxP

)(]0Pr[]0Pr[)( xFXXxF YX

p

pFxP Y 1

1

)(xFY

Page 6: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

The general model is as follow:

so, we actually build inverse conditional distribution function:

where А is all possible

As it is more convenient to take logarithmic function

Then receive:

A NEW APPROACH TO ESTIMATENET PREMIUM RATES

)(1 WβgY

),()( 11 Wβ gFY )1,0(

)(g )ln(

eeeeeY kkWW 11110 ...Wβ

Page 7: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

The idea of new approach is to estimate quantile regression of type

with parameter

- Quantile is seen as a good tool to separate good and bad risks, thus, the method proposed gives adequate unbiased estimates.

- The avoidance of dividing the problem of the estimation of net premium rates into two subtasks – the assessment of expected net premium and the evaluation of risk loadings

A NEW APPROACH TO ESTIMATENET PREMIUM RATES

)()(1 WβgFY).(

Page 8: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

Conditional premium rate calculated by the formula:

as a function , it is sufficient to take the expectation:

Then premium rate for the whole portfolio will be calculated with the following formula:

A NEW APPROACH TO ESTIMATENET PREMIUM RATES

))(( 1 Wβ gPY

)(

))(( 1Wβ gEPY

.)1()1(0 YY PpPppP

Page 9: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

Analyzed data provided by one of the insurance companies operating on the motor insurance market in St. Petersburg (theft of vehicles): all contracts = 11790, where claims = 2359.

Let so

Characteristics of the object insured: the type of car, the colour, the region of theft and the date of theft.

-it is important to describe dependence of loss (claim size) from the characteristics of the object insured.

EXAMPLE

95.0 75.02.0

8.095.0

1

p

p

Page 10: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

EXAMPLEHistogram (CrossSection.sta 14v*2359c)

1600,000014504,3478

27408,695740313,0000

53217,391366121,7391

79026,087091930,4348

1,0483E51,1774E5

1,3064E51,4355E5

Cena

0

100

200

300

400

500

600

700

800

900N

o o

f o

bs

Histogram of losses.

Page 11: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

Assumptions about theoretical distribution function of losses:

1. Losses are LogNormally distributed.

2. Losses are Gamma-distributed.

-for both, the net premium estimates were achieved as a third quartile for appropriate constructed theoretical distribution.

By the method introduced in the paper, the model should be evaluated by quantile regression with parameter

-in the case of the method of quantile regression, it is enough to take expectation of the theoretical distribution.

EXAMPLE

75.0

Page 12: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

EXAMPLE

The results of estimating by all methods, are shown in the table:

Page 13: European University at St. Petersburg THE METHOD OF QUANTILE REGRESSION, A NEW APPROACH TO ACTUARIAL MATHEMATICS Authors: Ruslan Abduramanov Andrey Kudryavtsev

• The new method is free from some important assumptions which are critical for classical methods.

• The method proposed is based on quantiles which are popular for risk assessment in finance.

• The method proposed gives greater confidence to the actuary in the results received.

• The working capability and power of the method is shown with practical example on the real data.

CONCLUSION